This paper gives an overview of recent seismological evidence on heterogeneity in the upper mantle. While there is a growing consensus among seismologists on the low‐order spherical harmonic expansion of upper mantle structure, short‐ and intermediate‐wavelength features remain the subject of intense debate. Regional studies of Earth structure yield evidence that the upper mantle is heterogeous at all resolvable scales and that global models underestimate the amplitudes of velocity anomalies. Some important problems still outstanding concern: (1) the fate of subducting slabs, (2) the depth extent of low‐velocity zones under spreading centers and (3) the depth at which hotspots originate. The gross radial layering of the upper mantle has become much more evident to us in recent years, mainly as a result of the analysis of waveforms observed by the modern generation of digital, broadband stations and by densely spaced, narrow band arrays of seismometers. Nevertheless, important questions on the average radial structure still persist. The low shear velocity zone has been mapped in detail beneath the continents, and Jordan's tectosphere hypothesis is largely confirmed by regional one‐dimensional (1‐D) and global tomographic studies. However, it is less clear to what extent these low shear velocities are matched by low compressional velocities. Anisotropy, for which compelling evidence exists in the observations of split shear waves, may play a role in shaping our images of low velocity zones. Below the asthenosphere, the existence of the Lehmann discontinuity at about 220 km depth as a worldwide feature is unlikely, but early indications for a phase transition at 520 km have found some confirmation in stacked wave sections. Anomalously low velocities at greater depth have been observed near current subduction zones and even beneath more ancient suture zones. Recent progress in analysis of seismic waves has followed major advances in seismic instrumentation of two kinds: (1) the installation of very broadband stations around the world, and (2) the increased density of stations in seismic networks. Future progress will most likely depend on the expansion of dense networks, including temporary and portable stations; on our ability to increase the station coverage of oceanic regions; and on advancing the forward problem of computing waveforms for heterogeneous Earth models to a level of efficiency that allows for nonlinear optimization.